forked from lijiext/lammps
git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@5541 f3b2605a-c512-4ea7-a41b-209d697bcdaa
This commit is contained in:
parent
52e8782482
commit
c837659cdb
|
@ -24,6 +24,11 @@ if (test $1 = 1) then
|
|||
cp pair_lj_cut_coul_long_gpu.h ..
|
||||
fi
|
||||
|
||||
if (test -e ../pair_lj_charmm_coul_long.cpp) then
|
||||
cp pair_lj_charmm_coul_long_gpu.cpp ..
|
||||
cp pair_lj_charmm_coul_long_gpu.h ..
|
||||
fi
|
||||
|
||||
if (test -e ../pair_cg_cmm.cpp) then
|
||||
cp pair_cg_cmm_gpu.cpp ..
|
||||
cp pair_cg_cmm_gpu.h ..
|
||||
|
@ -56,6 +61,7 @@ elif (test $1 = 0) then
|
|||
rm ../pair_lj96_cut_gpu.cpp
|
||||
rm ../pair_lj_cut_coul_cut_gpu.cpp
|
||||
rm ../pair_lj_cut_coul_long_gpu.cpp
|
||||
rm ../pair_lj_charmm_coul_long_gpu.cpp
|
||||
rm ../pair_cg_cmm_gpu.cpp
|
||||
rm ../pair_cg_cmm_coul_long_gpu.cpp
|
||||
rm ../fix_gpu.cpp
|
||||
|
@ -65,8 +71,10 @@ elif (test $1 = 0) then
|
|||
rm ../pair_lj96_cut_gpu.h
|
||||
rm ../pair_lj_cut_coul_cut_gpu.h
|
||||
rm ../pair_lj_cut_coul_long_gpu.h
|
||||
rm ../pair_lj_charmm_coul_long_gpu.h
|
||||
rm ../pair_cg_cmm_gpu.h
|
||||
rm ../pair_cg_cmm_coul_long_gpu.h
|
||||
rm ../fix_gpu.h
|
||||
|
||||
fi
|
||||
|
||||
|
|
|
@ -23,13 +23,16 @@
|
|||
#include "timer.h"
|
||||
#include "modify.h"
|
||||
#include "domain.h"
|
||||
#include "universe.h"
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
enum{GPU_FORCE, GPU_NEIGH};
|
||||
|
||||
extern bool lmp_init_device(const int first_gpu, const int last_gpu,
|
||||
const int gpu_mode, const double particle_split);
|
||||
extern bool lmp_init_device(MPI_Comm world, MPI_Comm replica,
|
||||
const int first_gpu, const int last_gpu,
|
||||
const int gpu_mode, const double particle_split,
|
||||
const int nthreads);
|
||||
extern void lmp_clear_device();
|
||||
extern double lmp_gpu_forces(double **f, double **tor, double *eatom,
|
||||
double **vatom, double *virial, double &ecoul);
|
||||
|
@ -39,7 +42,7 @@ extern double lmp_gpu_forces(double **f, double **tor, double *eatom,
|
|||
FixGPU::FixGPU(LAMMPS *lmp, int narg, char **arg) :
|
||||
Fix(lmp, narg, arg)
|
||||
{
|
||||
if (narg < 7) error->all("Illegal fix gpu command");
|
||||
if (narg != 7) error->all("Illegal fix gpu command");
|
||||
|
||||
if (strcmp(arg[1],"all") != 0)
|
||||
error->all("Illegal fix gpu command");
|
||||
|
@ -62,8 +65,9 @@ FixGPU::FixGPU(LAMMPS *lmp, int narg, char **arg) :
|
|||
particle_split = force->numeric(arg[6]);
|
||||
if (particle_split==0 || particle_split>1)
|
||||
error->all("Illegal fix gpu command.");
|
||||
|
||||
if (!lmp_init_device(first_gpu,last_gpu,gpu_mode,particle_split))
|
||||
|
||||
if (!lmp_init_device(universe->uworld,world,first_gpu,last_gpu,gpu_mode,
|
||||
particle_split,1))
|
||||
error->one("Could not find or initialize a specified accelerator device.");
|
||||
}
|
||||
|
||||
|
|
|
@ -95,9 +95,7 @@ PairCGCMMCoulLongGPU::~PairCGCMMCoulLongGPU()
|
|||
|
||||
void PairCGCMMCoulLongGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (update->ntimestep > MAXSMALLINT)
|
||||
error->all("Timestep too big for GPU pair style");
|
||||
int ntimestep = update->ntimestep;
|
||||
int ntimestep = static_cast<int>(update->ntimestep % MAXSMALLINT);
|
||||
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
|
|
@ -84,9 +84,7 @@ PairCGCMMGPU::~PairCGCMMGPU()
|
|||
|
||||
void PairCGCMMGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (update->ntimestep > MAXSMALLINT)
|
||||
error->all("Timestep too big for GPU pair style");
|
||||
int ntimestep = update->ntimestep;
|
||||
int ntimestep = static_cast<int>(update->ntimestep % MAXSMALLINT);
|
||||
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
|
|
@ -89,9 +89,7 @@ PairGayBerneGPU::~PairGayBerneGPU()
|
|||
|
||||
void PairGayBerneGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (update->ntimestep > MAXSMALLINT)
|
||||
error->all("Timestep too big for GPU pair style");
|
||||
int ntimestep = update->ntimestep;
|
||||
int ntimestep = static_cast<int>(update->ntimestep % MAXSMALLINT);
|
||||
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
|
|
@ -83,9 +83,7 @@ PairLJ96CutGPU::~PairLJ96CutGPU()
|
|||
|
||||
void PairLJ96CutGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (update->ntimestep > MAXSMALLINT)
|
||||
error->all("Timestep too big for GPU pair style");
|
||||
int ntimestep = update->ntimestep;
|
||||
int ntimestep = static_cast<int>(update->ntimestep % MAXSMALLINT);
|
||||
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
|
|
@ -0,0 +1,488 @@
|
|||
/* ----------------------------------------------------------------------
|
||||
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
|
||||
http://lammps.sandia.gov, Sandia National Laboratories
|
||||
Steve Plimpton, sjplimp@sandia.gov
|
||||
|
||||
Copyright (2003) Sandia Corporation. Under the terms of Contract
|
||||
DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
|
||||
certain rights in this software. This software is distributed under
|
||||
the GNU General Public License.
|
||||
|
||||
See the README file in the top-level LAMMPS directory.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
Contributing author: Mike Brown (SNL)
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#include "math.h"
|
||||
#include "stdio.h"
|
||||
#include "stdlib.h"
|
||||
#include "pair_lj_charmm_coul_long_gpu.h"
|
||||
#include "lmptype.h"
|
||||
#include "atom.h"
|
||||
#include "atom_vec.h"
|
||||
#include "comm.h"
|
||||
#include "force.h"
|
||||
#include "neighbor.h"
|
||||
#include "neigh_list.h"
|
||||
#include "integrate.h"
|
||||
#include "memory.h"
|
||||
#include "error.h"
|
||||
#include "neigh_request.h"
|
||||
#include "universe.h"
|
||||
#include "update.h"
|
||||
#include "domain.h"
|
||||
#include "string.h"
|
||||
#include "kspace.h"
|
||||
|
||||
#define MIN(a,b) ((a) < (b) ? (a) : (b))
|
||||
#define MAX(a,b) ((a) > (b) ? (a) : (b))
|
||||
|
||||
#define EWALD_F 1.12837917
|
||||
#define EWALD_P 0.3275911
|
||||
#define A1 0.254829592
|
||||
#define A2 -0.284496736
|
||||
#define A3 1.421413741
|
||||
#define A4 -1.453152027
|
||||
#define A5 1.061405429
|
||||
|
||||
// External functions from cuda library for atom decomposition
|
||||
|
||||
bool crml_gpu_init(const int ntypes, double cut_bothsq, double **host_lj1,
|
||||
double **host_lj2, double **host_lj3, double **host_lj4,
|
||||
double **offset, double *special_lj, const int nlocal,
|
||||
const int nall, const int max_nbors, const int maxspecial,
|
||||
const double cell_size, int &gpu_mode, FILE *screen,
|
||||
double host_cut_ljsq, double host_cut_coulsq,
|
||||
double *host_special_coul, const double qqrd2e,
|
||||
const double g_ewald, const double cut_lj_innersq,
|
||||
const double denom_lj, double **epsilon, double **sigma,
|
||||
const bool mix_arithmetic);
|
||||
void crml_gpu_clear();
|
||||
int * crml_gpu_compute_n(const int timestep, const int ago, const int inum,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
double *boxlo, double *boxhi, int *tag, int **nspecial,
|
||||
int **special, const bool eflag, const bool vflag,
|
||||
const bool eatom, const bool vatom, int &host_start,
|
||||
const double cpu_time, bool &success, double *host_q);
|
||||
void crml_gpu_compute(const int timestep, const int ago, const int inum,
|
||||
const int nall, double **host_x, int *host_type,
|
||||
int *ilist, int *numj, int **firstneigh,
|
||||
const bool eflag, const bool vflag, const bool eatom,
|
||||
const bool vatom, int &host_start, const double cpu_time,
|
||||
bool &success, double *host_q);
|
||||
double crml_gpu_bytes();
|
||||
|
||||
using namespace LAMMPS_NS;
|
||||
|
||||
enum{GEOMETRIC,ARITHMETIC,SIXTHPOWER};
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairLJCharmmCoulLongGPU::PairLJCharmmCoulLongGPU(LAMMPS *lmp) :
|
||||
PairLJCharmmCoulLong(lmp), gpu_mode(GPU_PAIR)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
free all arrays
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
PairLJCharmmCoulLongGPU::~PairLJCharmmCoulLongGPU()
|
||||
{
|
||||
crml_gpu_clear();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairLJCharmmCoulLongGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
int ntimestep = static_cast<int>(update->ntimestep % MAXSMALLINT);
|
||||
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
||||
int nall = atom->nlocal + atom->nghost;
|
||||
int inum, host_start;
|
||||
|
||||
bool success = true;
|
||||
|
||||
if (gpu_mode == GPU_NEIGH) {
|
||||
inum = atom->nlocal;
|
||||
gpulist = crml_gpu_compute_n(ntimestep, neighbor->ago, inum, nall,
|
||||
atom->x, atom->type, domain->sublo,
|
||||
domain->subhi, atom->tag, atom->nspecial,
|
||||
atom->special, eflag, vflag, eflag_atom,
|
||||
vflag_atom, host_start, cpu_time, success,
|
||||
atom->q);
|
||||
} else {
|
||||
inum = list->inum;
|
||||
crml_gpu_compute(ntimestep, neighbor->ago, inum, nall, atom->x,
|
||||
atom->type, list->ilist, list->numneigh, list->firstneigh,
|
||||
eflag, vflag, eflag_atom, vflag_atom, host_start, cpu_time,
|
||||
success, atom->q);
|
||||
}
|
||||
if (!success)
|
||||
error->one("Out of memory on GPGPU");
|
||||
|
||||
if (host_start<inum) {
|
||||
cpu_time = MPI_Wtime();
|
||||
if (gpu_mode == GPU_NEIGH)
|
||||
cpu_compute(gpulist, host_start, eflag, vflag);
|
||||
else
|
||||
cpu_compute(host_start, eflag, vflag);
|
||||
cpu_time = MPI_Wtime() - cpu_time;
|
||||
}
|
||||
}
|
||||
|
||||
/* ----------------------------------------------------------------------
|
||||
init specific to this pair style
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
void PairLJCharmmCoulLongGPU::init_style()
|
||||
{
|
||||
cut_respa = NULL;
|
||||
|
||||
if (!atom->q_flag)
|
||||
error->all("Pair style lj/charmm/coul/long requires atom attribute q");
|
||||
if (force->pair_match("gpu",0) == NULL)
|
||||
error->all("Cannot use pair hybrid with multiple GPU pair styles");
|
||||
|
||||
// Repeat cutsq calculation because done after call to init_style
|
||||
double cut;
|
||||
for (int i = 1; i <= atom->ntypes; i++) {
|
||||
for (int j = i; j <= atom->ntypes; j++) {
|
||||
if (setflag[i][j] != 0 || (setflag[i][i] != 0 && setflag[j][j] != 0))
|
||||
cut = init_one(i,j);
|
||||
}
|
||||
}
|
||||
|
||||
cut_lj_innersq = cut_lj_inner * cut_lj_inner;
|
||||
cut_ljsq = cut_lj * cut_lj;
|
||||
cut_coulsq = cut_coul * cut_coul;
|
||||
cut_bothsq = MAX(cut_ljsq,cut_coulsq);
|
||||
|
||||
denom_lj = (cut_ljsq-cut_lj_innersq) * (cut_ljsq-cut_lj_innersq) *
|
||||
(cut_ljsq-cut_lj_innersq);
|
||||
|
||||
double cell_size = sqrt(cut_bothsq) + neighbor->skin;
|
||||
|
||||
// insure use of KSpace long-range solver, set g_ewald
|
||||
|
||||
if (force->kspace == NULL)
|
||||
error->all("Pair style is incompatible with KSpace style");
|
||||
g_ewald = force->kspace->g_ewald;
|
||||
|
||||
// setup force tables
|
||||
|
||||
if (ncoultablebits) init_tables();
|
||||
|
||||
int maxspecial=0;
|
||||
if (atom->molecular)
|
||||
maxspecial=atom->maxspecial;
|
||||
bool init_ok = crml_gpu_init(atom->ntypes+1, cut_bothsq, lj1, lj2, lj3, lj4,
|
||||
offset, force->special_lj, atom->nlocal,
|
||||
atom->nlocal+atom->nghost, 300, maxspecial,
|
||||
cell_size, gpu_mode, screen, cut_ljsq,
|
||||
cut_coulsq, force->special_coul, force->qqrd2e,
|
||||
g_ewald, cut_lj_innersq,denom_lj,epsilon,sigma,
|
||||
mix_flag == ARITHMETIC);
|
||||
if (!init_ok)
|
||||
error->one("Insufficient memory on accelerator (or no fix gpu).\n");
|
||||
|
||||
if (force->newton_pair)
|
||||
error->all("Cannot use newton pair with GPU CHARMM pair style");
|
||||
|
||||
if (gpu_mode != GPU_NEIGH) {
|
||||
int irequest = neighbor->request(this);
|
||||
neighbor->requests[irequest]->half = 0;
|
||||
neighbor->requests[irequest]->full = 1;
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
double PairLJCharmmCoulLongGPU::memory_usage()
|
||||
{
|
||||
double bytes = Pair::memory_usage();
|
||||
return bytes + crml_gpu_bytes();
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairLJCharmmCoulLongGPU::cpu_compute(int start, int eflag, int vflag)
|
||||
{
|
||||
int i,j,ii,jj,inum,jnum,itype,jtype,itable;
|
||||
double qtmp,xtmp,ytmp,ztmp,delx,dely,delz,evdwl,ecoul,fpair;
|
||||
double fraction,table;
|
||||
double r,r2inv,r6inv,forcecoul,forcelj,factor_coul,factor_lj;
|
||||
double grij,expm2,prefactor,t,erfc;
|
||||
double philj,switch1,switch2;
|
||||
int *ilist,*jlist,*numneigh,**firstneigh;
|
||||
double rsq;
|
||||
|
||||
evdwl = ecoul = 0.0;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
double *q = atom->q;
|
||||
int *type = atom->type;
|
||||
int nlocal = atom->nlocal;
|
||||
int nall = nlocal + atom->nghost;
|
||||
double *special_coul = force->special_coul;
|
||||
double *special_lj = force->special_lj;
|
||||
double qqrd2e = force->qqrd2e;
|
||||
|
||||
inum = list->inum;
|
||||
ilist = list->ilist;
|
||||
numneigh = list->numneigh;
|
||||
firstneigh = list->firstneigh;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (ii = start; ii < inum; ii++) {
|
||||
i = ilist[ii];
|
||||
qtmp = q[i];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
jlist = firstneigh[i];
|
||||
jnum = numneigh[i];
|
||||
|
||||
for (jj = 0; jj < jnum; jj++) {
|
||||
j = jlist[jj];
|
||||
|
||||
if (j < nall) factor_coul = factor_lj = 1.0;
|
||||
else {
|
||||
factor_coul = special_coul[j/nall];
|
||||
factor_lj = special_lj[j/nall];
|
||||
j %= nall;
|
||||
}
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
|
||||
if (rsq < cut_bothsq) {
|
||||
r2inv = 1.0/rsq;
|
||||
|
||||
if (rsq < cut_coulsq) {
|
||||
if (!ncoultablebits || rsq <= tabinnersq) {
|
||||
r = sqrt(rsq);
|
||||
grij = g_ewald * r;
|
||||
expm2 = exp(-grij*grij);
|
||||
t = 1.0 / (1.0 + EWALD_P*grij);
|
||||
erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;
|
||||
prefactor = qqrd2e * qtmp*q[j]/r;
|
||||
forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);
|
||||
if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else {
|
||||
union_int_float_t rsq_lookup;
|
||||
rsq_lookup.f = rsq;
|
||||
itable = rsq_lookup.i & ncoulmask;
|
||||
itable >>= ncoulshiftbits;
|
||||
fraction = (rsq_lookup.f - rtable[itable]) * drtable[itable];
|
||||
table = ftable[itable] + fraction*dftable[itable];
|
||||
forcecoul = qtmp*q[j] * table;
|
||||
if (factor_coul < 1.0) {
|
||||
table = ctable[itable] + fraction*dctable[itable];
|
||||
prefactor = qtmp*q[j] * table;
|
||||
forcecoul -= (1.0-factor_coul)*prefactor;
|
||||
}
|
||||
}
|
||||
} else forcecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq) {
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
jtype = type[j];
|
||||
forcelj = r6inv * (lj1[itype][jtype]*r6inv - lj2[itype][jtype]);
|
||||
if (rsq > cut_lj_innersq) {
|
||||
switch1 = (cut_ljsq-rsq) * (cut_ljsq-rsq) *
|
||||
(cut_ljsq + 2.0*rsq - 3.0*cut_lj_innersq) / denom_lj;
|
||||
switch2 = 12.0*rsq * (cut_ljsq-rsq) *
|
||||
(rsq-cut_lj_innersq) / denom_lj;
|
||||
philj = r6inv * (lj3[itype][jtype]*r6inv - lj4[itype][jtype]);
|
||||
forcelj = forcelj*switch1 + philj*switch2;
|
||||
}
|
||||
} else forcelj = 0.0;
|
||||
|
||||
fpair = (forcecoul + factor_lj*forcelj) * r2inv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
if (rsq < cut_coulsq) {
|
||||
if (!ncoultablebits || rsq <= tabinnersq)
|
||||
ecoul = prefactor*erfc;
|
||||
else {
|
||||
table = etable[itable] + fraction*detable[itable];
|
||||
ecoul = qtmp*q[j] * table;
|
||||
}
|
||||
if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else ecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq) {
|
||||
evdwl = r6inv*(lj3[itype][jtype]*r6inv-lj4[itype][jtype]);
|
||||
if (rsq > cut_lj_innersq) {
|
||||
switch1 = (cut_ljsq-rsq) * (cut_ljsq-rsq) *
|
||||
(cut_ljsq + 2.0*rsq - 3.0*cut_lj_innersq) / denom_lj;
|
||||
evdwl *= switch1;
|
||||
}
|
||||
evdwl *= factor_lj;
|
||||
} else evdwl = 0.0;
|
||||
}
|
||||
|
||||
if (evflag) ev_tally_full(i,evdwl,ecoul,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
void PairLJCharmmCoulLongGPU::cpu_compute(int *nbors, int start, int eflag,
|
||||
int vflag)
|
||||
{
|
||||
int i,j,jnum,itype,jtype,itable;
|
||||
double qtmp,xtmp,ytmp,ztmp,delx,dely,delz,evdwl,ecoul,fpair;
|
||||
double fraction,table;
|
||||
double r,r2inv,r6inv,forcecoul,forcelj,factor_coul,factor_lj;
|
||||
double grij,expm2,prefactor,t,erfc;
|
||||
double philj,switch1,switch2;
|
||||
double rsq;
|
||||
|
||||
evdwl = ecoul = 0.0;
|
||||
|
||||
double **x = atom->x;
|
||||
double **f = atom->f;
|
||||
double *q = atom->q;
|
||||
int *type = atom->type;
|
||||
int nlocal = atom->nlocal;
|
||||
int nall = nlocal + atom->nghost;
|
||||
int stride = nlocal - start;
|
||||
double *special_coul = force->special_coul;
|
||||
double *special_lj = force->special_lj;
|
||||
double qqrd2e = force->qqrd2e;
|
||||
|
||||
// loop over neighbors of my atoms
|
||||
|
||||
for (i = start; i < nlocal; i++) {
|
||||
qtmp = q[i];
|
||||
xtmp = x[i][0];
|
||||
ytmp = x[i][1];
|
||||
ztmp = x[i][2];
|
||||
itype = type[i];
|
||||
int *nbor = nbors + i - start;
|
||||
jnum = *nbor;
|
||||
nbor += stride;
|
||||
int *nbor_end = nbor + stride * jnum;
|
||||
|
||||
for (; nbor<nbor_end; nbor+=stride) {
|
||||
j = *nbor;
|
||||
|
||||
if (j < nall) factor_coul = factor_lj = 1.0;
|
||||
else {
|
||||
factor_coul = special_coul[j/nall];
|
||||
factor_lj = special_lj[j/nall];
|
||||
j %= nall;
|
||||
}
|
||||
|
||||
delx = xtmp - x[j][0];
|
||||
dely = ytmp - x[j][1];
|
||||
delz = ztmp - x[j][2];
|
||||
rsq = delx*delx + dely*dely + delz*delz;
|
||||
|
||||
if (rsq < cut_bothsq) {
|
||||
r2inv = 1.0/rsq;
|
||||
|
||||
if (rsq < cut_coulsq) {
|
||||
if (!ncoultablebits || rsq <= tabinnersq) {
|
||||
r = sqrt(rsq);
|
||||
grij = g_ewald * r;
|
||||
expm2 = exp(-grij*grij);
|
||||
t = 1.0 / (1.0 + EWALD_P*grij);
|
||||
erfc = t * (A1+t*(A2+t*(A3+t*(A4+t*A5)))) * expm2;
|
||||
prefactor = qqrd2e * qtmp*q[j]/r;
|
||||
forcecoul = prefactor * (erfc + EWALD_F*grij*expm2);
|
||||
if (factor_coul < 1.0) forcecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else {
|
||||
union_int_float_t rsq_lookup;
|
||||
rsq_lookup.f = rsq;
|
||||
itable = rsq_lookup.i & ncoulmask;
|
||||
itable >>= ncoulshiftbits;
|
||||
fraction = (rsq_lookup.f - rtable[itable]) * drtable[itable];
|
||||
table = ftable[itable] + fraction*dftable[itable];
|
||||
forcecoul = qtmp*q[j] * table;
|
||||
if (factor_coul < 1.0) {
|
||||
table = ctable[itable] + fraction*dctable[itable];
|
||||
prefactor = qtmp*q[j] * table;
|
||||
forcecoul -= (1.0-factor_coul)*prefactor;
|
||||
}
|
||||
}
|
||||
} else forcecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq) {
|
||||
r6inv = r2inv*r2inv*r2inv;
|
||||
jtype = type[j];
|
||||
forcelj = r6inv * (lj1[itype][jtype]*r6inv - lj2[itype][jtype]);
|
||||
if (rsq > cut_lj_innersq) {
|
||||
switch1 = (cut_ljsq-rsq) * (cut_ljsq-rsq) *
|
||||
(cut_ljsq + 2.0*rsq - 3.0*cut_lj_innersq) / denom_lj;
|
||||
switch2 = 12.0*rsq * (cut_ljsq-rsq) *
|
||||
(rsq-cut_lj_innersq) / denom_lj;
|
||||
philj = r6inv * (lj3[itype][jtype]*r6inv - lj4[itype][jtype]);
|
||||
forcelj = forcelj*switch1 + philj*switch2;
|
||||
}
|
||||
} else forcelj = 0.0;
|
||||
|
||||
fpair = (forcecoul + factor_lj*forcelj) * r2inv;
|
||||
|
||||
f[i][0] += delx*fpair;
|
||||
f[i][1] += dely*fpair;
|
||||
f[i][2] += delz*fpair;
|
||||
|
||||
if (eflag) {
|
||||
if (rsq < cut_coulsq) {
|
||||
if (!ncoultablebits || rsq <= tabinnersq)
|
||||
ecoul = prefactor*erfc;
|
||||
else {
|
||||
table = etable[itable] + fraction*detable[itable];
|
||||
ecoul = qtmp*q[j] * table;
|
||||
}
|
||||
if (factor_coul < 1.0) ecoul -= (1.0-factor_coul)*prefactor;
|
||||
} else ecoul = 0.0;
|
||||
|
||||
if (rsq < cut_ljsq) {
|
||||
evdwl = r6inv*(lj3[itype][jtype]*r6inv-lj4[itype][jtype]);
|
||||
if (rsq > cut_lj_innersq) {
|
||||
switch1 = (cut_ljsq-rsq) * (cut_ljsq-rsq) *
|
||||
(cut_ljsq + 2.0*rsq - 3.0*cut_lj_innersq) / denom_lj;
|
||||
evdwl *= switch1;
|
||||
}
|
||||
evdwl *= factor_lj;
|
||||
} else evdwl = 0.0;
|
||||
}
|
||||
|
||||
if (j<start) {
|
||||
if (evflag) ev_tally_full(i,evdwl,ecoul,fpair,delx,dely,delz);
|
||||
} else {
|
||||
if (j<nlocal) {
|
||||
f[j][0] -= delx*fpair;
|
||||
f[j][1] -= dely*fpair;
|
||||
f[j][2] -= delz*fpair;
|
||||
}
|
||||
if (evflag) ev_tally(i,j,nlocal,0,
|
||||
evdwl,ecoul,fpair,delx,dely,delz);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -0,0 +1,48 @@
|
|||
/* ----------------------------------------------------------------------
|
||||
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
|
||||
http://lammps.sandia.gov, Sandia National Laboratories
|
||||
Steve Plimpton, sjplimp@sandia.gov
|
||||
|
||||
Copyright (2003) Sandia Corporation. Under the terms of Contract
|
||||
DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
|
||||
certain rights in this software. This software is distributed under
|
||||
the GNU General Public License.
|
||||
|
||||
See the README file in the top-level LAMMPS directory.
|
||||
------------------------------------------------------------------------- */
|
||||
|
||||
#ifdef PAIR_CLASS
|
||||
|
||||
PairStyle(lj/charmm/coul/long/gpu,PairLJCharmmCoulLongGPU)
|
||||
|
||||
#else
|
||||
|
||||
#ifndef LMP_PAIR_LJ_CHARMM_COUL_LONG_GPU_H
|
||||
#define LMP_PAIR_LJ_CHARMM_COUL_LONG_GPU_H
|
||||
|
||||
#include "pair_lj_charmm_coul_long.h"
|
||||
|
||||
namespace LAMMPS_NS {
|
||||
|
||||
class PairLJCharmmCoulLongGPU : public PairLJCharmmCoulLong {
|
||||
public:
|
||||
PairLJCharmmCoulLongGPU(LAMMPS *lmp);
|
||||
~PairLJCharmmCoulLongGPU();
|
||||
void cpu_compute(int, int, int);
|
||||
void cpu_compute(int *, int, int, int);
|
||||
void compute(int, int);
|
||||
void init_style();
|
||||
double memory_usage();
|
||||
|
||||
enum { GPU_PAIR, GPU_NEIGH };
|
||||
|
||||
private:
|
||||
int gpu_mode;
|
||||
double cpu_time;
|
||||
int *gpulist;
|
||||
};
|
||||
|
||||
}
|
||||
#endif
|
||||
#endif
|
||||
|
|
@ -85,9 +85,7 @@ PairLJCutCoulCutGPU::~PairLJCutCoulCutGPU()
|
|||
|
||||
void PairLJCutCoulCutGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (update->ntimestep > MAXSMALLINT)
|
||||
error->all("Timestep too big for GPU pair style");
|
||||
int ntimestep = update->ntimestep;
|
||||
int ntimestep = static_cast<int>(update->ntimestep % MAXSMALLINT);
|
||||
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
|
|
@ -76,7 +76,8 @@ using namespace LAMMPS_NS;
|
|||
|
||||
/* ---------------------------------------------------------------------- */
|
||||
|
||||
PairLJCutCoulLongGPU::PairLJCutCoulLongGPU(LAMMPS *lmp) : PairLJCutCoulLong(lmp), gpu_mode(GPU_PAIR)
|
||||
PairLJCutCoulLongGPU::PairLJCutCoulLongGPU(LAMMPS *lmp) :
|
||||
PairLJCutCoulLong(lmp), gpu_mode(GPU_PAIR)
|
||||
{
|
||||
respa_enable = 0;
|
||||
cpu_time = 0.0;
|
||||
|
@ -95,9 +96,7 @@ PairLJCutCoulLongGPU::~PairLJCutCoulLongGPU()
|
|||
|
||||
void PairLJCutCoulLongGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (update->ntimestep > MAXSMALLINT)
|
||||
error->all("Timestep too big for GPU pair style");
|
||||
int ntimestep = update->ntimestep;
|
||||
int ntimestep = static_cast<int>(update->ntimestep % MAXSMALLINT);
|
||||
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
|
|
@ -83,9 +83,7 @@ PairLJCutGPU::~PairLJCutGPU()
|
|||
|
||||
void PairLJCutGPU::compute(int eflag, int vflag)
|
||||
{
|
||||
if (update->ntimestep > MAXSMALLINT)
|
||||
error->all("Timestep too big for GPU pair style");
|
||||
int ntimestep = update->ntimestep;
|
||||
int ntimestep = static_cast<int>(update->ntimestep % MAXSMALLINT);
|
||||
|
||||
if (eflag || vflag) ev_setup(eflag,vflag);
|
||||
else evflag = vflag_fdotr = 0;
|
||||
|
|
Loading…
Reference in New Issue